Professional Documents
Culture Documents
Nafems Dynamic Fea Webinar April 09 PDF
Nafems Dynamic Fea Webinar April 09 PDF
Webinar
Ladzinski Abbey
Collaboration Innovation Productivity - Quality
THE INTERNATIONAL ASSOCIATION
FOR THE ENGINEERING ANALYSIS
COMMUNITY
Matthew Ladzinski
NAFEMS
North America
Introduction to FETraining
Q and A
Details
Email : tony@fetraining.com
www.fetraining.com
NAFEMS. The International Association for the Engineering Analysis Community
Creating Awareness Delivering Education and Training Stimulating Standards Dynamics E-Learning Course V1.0 Page 12
Intro to FETraining: Training Solutions
Interactive DVD
Details
Email : tony@fetraining.com
www.fetraining.com
www.nafems.org/events/nafems/2009/e-dynamicfea/
Many problems facing designers and engineers are dynamic in nature. The
response of a structure cannot be simply assessed using static assumptions.
The nature of the problem may be to understand the resonant frequencies of
your design, so that key driving frequencies such as equipment rotational speed,
acoustic or external pressure frequencies, ground motion frequency content or
vehicle passing frequency.
Your design may face external driving forces from adjacent components; cams,
push rods, pistons or from vehicle input sources such as a bumpy road, wave
loading, air pressure or inertial forces.
Whatever the nature of the challenge, this objective of this course is to break
down the dynamic problem into clearly defined steps, give an overview of the
physics involved and show how to successfully implement practical solutions
using Finite Element Analysis.
In the current climate travel and training budgets are tight. To help you still meet
your training needs the following e-learning course has been developed to
complement the live class.
The e-learning course runs over a six week period with a single two hour
session per week.
Bulletin Boards and Email are used to keep in contact between sessions,
mentoring homework and allowing interchange between students.
E-learning classes are ideal for companies with a group of engineers requiring
training. E-learning classes can be provided to suit your needs and timescale.
Contact us to discuss your requirements.
We hope that small companies or individuals can now take part in the training
experience.
Damping
What level of damping should I use?
We tend to use both terms to mean the resonant frequency of oscillation , but
more accurately:
We expect mode 1 to dominate intuitively, but notice we have a twisting mode maybe a
very heavy athlete could just catch a corner of the board and put some twist in?
Intuitively we would expect all these to be important modes, but as yet we cant prove
that.
The next range of modes are 160 Hz and above and are very complex shapes.
Important question ; what is the range of input or driving frequencies? That will dictate
which modes are critical.
We can use that as the basis of the range of interest. We cant use the range directly
as that ignores the very complex interaction we may see between harmonics of the
system and other factors. So we typically take an upper bound of 1.5 or 2 times the
upper frequency.
The lower bound should go right down to the lowest frequency we find, as it is difficult
to provide a sensible cut off here.
So that gives us a set of modes of interest to investigate. We know a lot about the
dynamic characteristics of the cup now. The basic building block approach.
Next stage is to look at how the cup is excited by the dashboard the response
analysis
However excitation frequency may be very high and drives he structural response
At these frequencies the interchange between the two forms of energy is very easily
triggered by any external input which has the same driving frequency.
These unique or special frequencies were labeled Eigen Frequencies in the early 20th
Century, from the German word. Mode shapes are correspondingly called Eigen Vectors
The energy balance is calculated by considering the inertial and stiffness terms in
isolation. We can find all the structural frequencies and mode shapes this way.
m = mass (inertia)
b = damping (energy dissipation)
k = stiffness (restoring force)
p = applied force
u = displacement of mass
= velocity of mass
= acceleration of mass
mu&&(t ) + bu& (t ) + ku (t ) = p (t )
mu&&(t ) + bu& (t ) + ku (t ) = p (t )
u (t ) = A sin nt + B cos nt
This form defines the response as being HARMONIC, combinations
of sine and cosine shape responses with a resonant frequency of:
k
:
n = = n 2
m
The displacement response is indeterminate as we dont know any initial
conditions
u& (0)
u (t ) = sin nt + u (0) cos nt
n
DOF: 1 2 3 4
1 1
1 1 + 1 1
[K ] = k
1 1 + 1 1
1 1
2 1 1 0
[K ] = k
and [M ] = m We lump masses
at DOF
1 2 0 2
Then 2 [M ]{ } + [K ]{ } = 0
This means we can find a mode
shape, {}, and frequency where
So ([K ] [M ]){ } = 0
2 the inertia terms and elastic terms
balance
2 1 2 1 0
1 2 m 0 2 { } = 0
k The Eigenvalue problem
NAFEMS. The International Association for the Engineering Analysis Community
Creating Awareness Delivering Education and Training Stimulating Standards Dynamics E-Learning Course V1.0 Page 35
What are Natural Frequencies, Normal Modes
For each of these sets, the inertia terms balance the elastic terms and
this is the definition of resonance
k 0.731
So at 1 = 0.634 , the motion is defined by: {1} =
m 1 .000
2.731
And at 2 = 2.366 k , the motion is defined by: { } =
1.000
2
m
k
Then 1 = 0.634 = 5.629 rads = 0.896 Hz
s
m
k
2 = 2.366 = 10.875 rads = 1.731Hz
s
m
Notice the conversion of Frequency from Radians/s to Cycles/s (Hertz)
f
=
2
NAFEMS. The International Association for the Engineering Analysis Community
Creating Awareness Delivering Education and Training Stimulating Standards Dynamics E-Learning Course V1.0 Page 38
What are Natural Frequencies, Normal Modes
Reduction techniques
Residual Vectors
Case Studies
The side effect of the Eigen Value method is that we do not know
the actual amplitude of the shapes that we calculate.
This confuses many users who are new to modal analysis.
0.731 1.000
{1} = {2 } =
1 . 000 0 .366
The simple answer is because we havent provided any excitation input or initial
conditions.
We dont know the weight of the diver or how high his spring was.
We dont know the speed of the car or the height of the bump.
All we know is what the range of natural frequencies of the board and the holder will be
and what their respective deflected shapes will look like.
But I see screen animations, they must have a value for deformation and they
look big?
The answer here is that we scale the magnitude of the shapes to one of several types
of arbitrary definitions for convenience and for shape comparison.
A post processor will further scale deformations so you can see each mode shape
clearly typically 10% of the maximum viewable dimension, again quite arbitrary.
The second part of the question is very important we are only dealing with linear,
small displacement theory. So in practice vibration amplitudes would have to be small
relative to the size of the structure.
NAFEMS. The International Association for the Engineering Analysis Community
Creating Awareness Delivering Education and Training Stimulating Standards Dynamics E-Learning Course V1.0 Page 42
Displacements out of Normal Modes Analysis
2144 Hz !!!
Test Analysis
Corresponding points on test and analysis model are located and deflections
measured, for each mode shape found
{ }Ti [M ]{ }i = 1 .0
{ }Ti [M ]{ }j = 0 .0
Diagonal { }TA [M ]{ }T = 1 .0
for perfect correlation
Off-Diagonal { }TA [M ]{ }T = 0 .0
It is the ratio between the diagonal and off diagonal terms which is
important
[{ } { } ]
T
A T
2
{ } { }T { }TT { }A
T
A
Project examples
Describing and assessing modes
Correlation Case studies
Damping
What level of damping should I use?
mu&&(t ) + bu& (t ) + ku (t ) = 0
There are 3 types of solution to this, defined as:
Critically Damped
Overdamped
Underdamped
A swing door with a dashpot closing mechanism is a good analogy:
If the door oscillates through the closed position it is underdamped
If it creeps slowly to the closed position it is overdamped
If it closes in the minimum possible time, with no overswing, it is critically
damped
For the critically damped case, there is no oscillation, just a decay from
the initial condition:
u (t ) = ( A + Bt )e bt / 2 m
The damping in this case is defined as:
b = bcr = 2 km = 2m n
A system is overdamped when b > bcr
For the underdamped case b < bcr and the solution is the form:
u (t ) = e bt / 2 m ( A sin d t + B cos d t )
d represents the Damped natural frequency of the system:
d = n 1 2
is called the Critical damping ratio and is defined by:
b
=
bcr
In most analyses is less than .1 (10%) so d n
Amplitude is a function
of damping
Disp.
5% Damping
2%
Damping
Time
NAFEMS. The International Association for the Engineering Analysis Community
Creating Awareness Delivering Education and Training Stimulating Standards Dynamics E-Learning Course V1.0 Page 64
Damping
I have discussed viscous damping but this is just one simulation method
Etc
The real world damping is a complex phenomena and is not fully understood
Testing and tuning against test results is the best approach
Now to answer the question, with a very rough rule of thumb and
using critical damping ratio:
2%
Damping
5% Damping
Bottom line is that a lower bound estimate is conservative, and upper bound is not
Always best to present a set of responses, based on two or three damping levels
NAFEMS. The International Association for the Engineering Analysis Community
Creating Awareness Delivering Education and Training Stimulating Standards Dynamics E-Learning Course V1.0 Page 67
Damping
{u&n } = 1
{un+1 un1} Disp Un+1 the unknown
2t u(t) un
{u&&n } = 1 2 {un+1 2un + un1}
t time
The time step chosen should be sufficiently small to capture the highest frequency
of interest in the response. For example, if this value is 100 Hz, each time period is
.01s (10ms) so we need at the very least 5 steps to capture the response, i.e.. t
=.002s (2ms). The preferred minimum number is 10 steps per period.
The accuracy of the load input is similarly dependent on the time step chosen, so a
loading with a 1000 Hz input will need at least t of .0002s (.2ms) and preferably
.0001s (.1ms)
Shock loading will have very high frequency content, either as an applied load, or
coming through a contact in non-linear
The smaller the value of t , the more accurate the integration will be, this may
override the previous comments.
Normally it will take a few runs to tune the model overall, and the effectiveness of
this technique can be investigated
The number of times steps needs to be adequate to ensure all over swings are
captured and that the response is decaying at the cut off point, with no surprises.
Reduce output to key nodes or elements to act as test points. Only when model
is thoroughly debugged then ask for full output
Loaded Free
d
G = 2 =
Where:
d1
d = ln
d2
Time steps are too coarse and misleading frequency content is seen
Rule of thumb
Explicit
Displacement
Time
Displacement
Frequency (Hz)
For >> 1
n
For 1
n
Note:
Use of a Log scale helps identify low
order response
Displacement
Frequency (Hz)
Must be at each natural frequency to ensure that the peak responses are captured
A general spread of points is required to capture the overall trend of the curve
The number of calculation points will increase CPU cost and output quantities
As with transient debug the model using key nodes and element responses before
running full output requests
Case studies
-g
Time
NAFEMS. The International Association for the Engineering Analysis Community
Creating Awareness Delivering Education and Training Stimulating Standards Dynamics E-Learning Course V1.0 Page 90
Quick peek at Shock Spectra and
Random analysis
Shock Spectra
A substructure or piece
of equipment with
frequency 3.5 Hz would
expect to see a peak
base acceleration of 1.4 g
Is the frequency range adequate (we will discuss this more in the section
on modal effective mass)
Plan a exploratory key point approach to the results, dont output all data
-g
Time
Q and A www.nafems.org
www.fetraining.com
Thank you!
matthew.ladzinski@nafems.org